Claude Sonnet 4.6 · WebArena(Official WebArena grader, modified fuzzy_match subgrader) 65.6% · 附件证据(ev46-webarena)

Claude Sonnet 4.6 的 WebArena · Official WebArena grader, modified fuzzy_match subgrader 官方出处。Evaluation settings not fully stated · 65.6%。见《Claude Sonnet 4.6 System Card》§2.18.1 Table 2.18.1.A, PDF p.38。证据 ev46-webarena。

原文与定位

Claude Sonnet 4.6 · WebArena(Official WebArena grader, modified fuzzy_match subgrader),Evaluation settings not fully stated配置,Average@5 score 65.6%(具体成绩公布时间未确认)。见《Claude Sonnet 4.6 System Card》§2.18.1 Table 2.18.1.A, PDF p.38。 查看成绩条件

Model Score Notes Claude Sonnet 4.6 65.6% Single policy model, general prompts Claude Opus 4.6 68.0% Single policy model, general prompts Claude Opus 4.5 65.3% Single policy model, general prompts Claude Sonnet 4.5 58.5% Single policy model, general prompts Claude Haiku 4.5 53.1% Single policy model, general prompts WebTactix 74.3% Multi-agent system OAgent 71.6% Multi-agent system OpenAI CUA 58.1% – [Table 2.18.1.A] WebArena performance.
官方材料
Claude Sonnet 4.6 System Card · PDF 第 38 页
定位
附件 · §2.18.1 Table 2.18.1.A, PDF p.38
材料版本
PDF with changelog dated 2026-03-06; 135 pages; collected 2026-11-10
版本公布
未确认
披露时间
未确认 · 首发/后补未确认
采集时间
历史依据
暂无可靠历史版本依据
引用文本

Claude Sonnet 4.6 · WebArena(Official WebArena grader, modified fuzzy_match subgrader),Evaluation settings not fully stated配置,Average@5 score 65.6%(具体成绩公布时间未确认)。官方出处 Claude Sonnet 4.6 System Card(附件,PDF with changelog dated 2026-03-06; 135 pages; collected 2026-11-10)§2.18.1 Table 2.18.1.A, PDF p.38,https://www.anthropic.com/claude-sonnet-4-6-system-card#page=38。Benchlog https://benchlog.lekshome.cc/evidence/claude-sonnet-4-6/ev46-webarena/

关联成绩与条件

Claude Sonnet 4.6 · WebArena 成绩与配置
指标与成绩配置与评测条件证据关系
Average@5 score65.6%原文 65.6% · 单位 %
公布 未确认
Evaluation settings not fully stated
设置
Only settings explicitly listed in each result's conditions are known.
未知项
Effort, sampling and tool settings are unspecified unless stated in result.conditions.
工具
允许工具
推理预算
未记录
框架
未记录
协议
Official WebArena grader; Average@5 over five independent runs.
条件
Single policy model, general prompts, Computer Use API with browser tools for screenshot and DOM-based navigation. Grader fuzzy_match base model changed from GPT-4 to Sonnet 4.5 and judge prompt rewritten.

直接支持该成绩

在排名中查看此成绩